A method and a virtual agent system services a user request from a user. The virtual agent system includes: (a) a conversational user interface receiving the user request and communicating with two or more virtual agents; and (b) a dialog manager including a natural language processing module, that directs operations of the conversational user interface, wherein the dialog manager (i) receives and analyzes the user request from the conversation user interface using the natural language processing module, (ii) causes the conversational user interface to request and to receive a response to the user request from each of the virtual agents, and (iii) integrates the received responses to the user request into an integrated response based on the natural language processing module and causes the conversational user interface to provide the integrated response to the user.
Legal claims defining the scope of protection, as filed with the USPTO.
1. A virtual agent system servicing a user request from a user, comprising: a conversational user interface receiving the user request and communicating with a plurality of virtual agents, wherein each virtual agent is associated with a profile created according to the virtual agent's content or service provided; and a dialog manager that directs operations of the conversational user interface, wherein the dialog manager, (i) receives the user request from the conversation user interface, (ii) causes the conversational user interface to select for collaboration a group of two or more virtual agents, based on an analysis of a context of the user request and the profile of each virtual agent selected, and to request and to receive a response to the user request from each of the virtual agents in the selected group, and (iii) integrates the received responses to the user request into an integrated response according to statistics specific to the virtual agents providing the received responses and causes the conversational user interface to provide the integrated response to the user.
2. The virtual agent system of claim I, wherein at least one of the virtual agents is tightly coupled to the dialog manager.
3. The virtual agent system of claim 1 , further comprising an interactive design console and wherein the at least one of the virtual agents that is tightly coupled to the dialog manager is trained using the interactive design console.
4. The virtual agent system of claim 1 , wherein at least one of the virtual agents is loosely-coupled to the dialog manager.
5. The virtual agent system of claim 4 , wherein the at least one of the virtual agents is coupled to the dialog manager through one or more application programming interfaces.
6. The virtual agent system of claim 1 , wherein the dialog manager comprises deep-learning models trained to perform the analysis of the user request.
7. The virtual agent system of claim 6 , wherein the deep-learning models are trained through semi-supervised techniques.
8. The virtual agent system of claim 1 , further comprising one or more of the following resources drawn on by the dialog manager: (a) a customer relationship database; (b) contact centers; and (c) additional virtual agents.
9. The virtual agent system of claim 1 , wherein each virtual agent communicates with the interactive user interface over a protocol which specifies one or more of: the virtual agent's content, conversation format, and result statistics.
10. The virtual agent system of claim 1 , wherein the virtual agents are selected by the user according to a browsing model.
11. The virtual agent system of claim 1 , wherein the selected group of virtual agents are selected according to a directory model, based on recognizing the context of the user request.
12. The virtual agent system of claim 11 , wherein the dialog manager generates a profile for each virtual agent, the profile comprising information or description regarding the content and services provided by the virtual agent.
13. The virtual agent system of claim 12 , wherein the selected group of virtual agents are selected by evaluating and ranking the groups of virtual agent based, at least in part, on applying a predetermined set of metrics on the recognized context.
14. The virtual agent system of claim 13 , wherein the metrics are recall-oriented.
15. The virtual agent system of claim 11 wherein, prior to grouping the virtual agents, the dialog manager causes the conversational user interface to send an initial request to a predetermined set of candidate virtual agents to solicit a potential response.
16. The virtual agent system of claim 15 , wherein the dialog manager integrates the potential response using precision-oriented statistics.
17. The virtual agent system of claim 16 , wherein the precision-oriented statistics comprise one or more of the following statistics: (a) inverse document frequencies, stop lists, historical requests, and context.
18. The virtual agent system of claim 15 , further comprising a sample response database, wherein the dialog manager queries the sample response database with the initial request.
19. The virtual agent system of claim 1 wherein the virtual agents are grouped according to one or more of the following criteria: (i) conversation clustering; (ii) relevant conversation distribution; (iii) rule-based selection; (iv) conversation probing; and (v) statistical.
20. The virtual agent system of claim 1 , wherein the virtual agents are grouped based on one or more of the following methods: (i) relevant request distribution and (ii) content-based bots ranking and selection.
21. The virtual agent system of claim 20 , wherein the relevant request distribution method comprises: (a) building a database using training data that associates user requests with relevant virtual agents; (b) for each new user request: (i) finding in the database k nearest “neighbors”; and (ii) obtaining a normalized relevant conversation distribution using statistics associated with the k neighbors to obtain a relevance metric for each potential virtual agent; and (c) selecting a desired number of federated conversational agents from the relevance metrics.
22. The virtual agent system of claim 1 , wherein the machine learning methods are used to adapt parameters of the dialog manager, which includes one or more of: domain, security, brand, past user responses, semantics, named entity features, category features, past user satisfaction features, vertical intent features, temporal features, text similarity features, hit count features, and other relevant parameters.
23. The virtual agent system of claim 1 , wherein the group of virtual agents are selected both according to the context and content the user request.
24. The virtual agent system of claim 1 , wherein the virtual agents are grouped according to a message-passing model implemented in a peer-to-peer network.
25. The virtual agent system of claim 24 , wherein each node of the peer-to-peer network comprises a client, a server, or both.
26. The virtual agent system of claim 24 , wherein the peer-to-peer network implements a flood algorithm to distribute the user request to the virtual agents.
27. The virtual agent system of claim 26 , wherein each virtual agent decides whether or not to respond to the user request.
28. The virtual agent system of claim 24 , wherein the peer-to-peer network has a central-hub topology.
29. The virtual agent system of claim 24 , wherein the peer-to-peer network has a hierarchical topology including hub nodes and leaf nodes.
30. The virtual agent system of claim 24 , wherein the peer-to-peer network implements one or more of: (a) partial indexing over specific network resources; (b) structured peer-to-peer architecture using distributed hash tables, (iii) a gossip protocol, (iv) an architecture organized according to content or user interests; (v) super peers; and (vi) reputation management.
31. In a virtual agent system servicing a user request from a user, a method comprising: in a conversational user interface, receiving the user request and communicating with a plurality of virtual agents, wherein each virtual agent is associated with a profile created according to the virtual agent's content or service provided; and in a dialog manager that directs operations of the conversational user interface: (i) receiving the user request from the conversation user interface, (ii) causing the conversational user interface to select for collaboration a group of two or more virtual agents. based on an analysis of a context of the user request and the profile of each virtual agent selected, and to request and to receive a response to the user request from each of the virtual agents in the selected group, and (iii) integrating the received responses to the user request into an integrated response according to statistics specific to the virtual agents providing the received responses and causes the conversational user interface to provide the integrated response to the user.
32. The method of claim 31 , wherein at east one of the virtual agents is tightly coupled to the dialog manager.
33. The method of claim 31 , further comprising providing an interactive design console and training the at least one of the virtual agents that is tightly coupled to the dialog manager using the interactive design console.
34. The method of claim 31 , wherein at least one of the virtual agents is loosely-coupled to the dialog manager.
35. The method of claim 34 , wherein the at least one of the virtual agents is coupled to the dialog manager through one or more application programming interfaces.
36. The method of claim 31 , wherein the dialog manager comprises deep-learning models trained to perform the analysis of the user request.
37. The method of claim 36 , further comprising training the deep- learning models using semi-supervised techniques.
38. The method of claim 31 , further comprising drawing on one or more of the following resources: (a) a customer relationship database; (b) contact centers; and (c) additional virtual agents.
39. The method of claim 31 , wherein each virtual agent communicates with the interactive user interface over a protocol which specifies one or more of: the virtual agent's content, conversation format, and result statistics.
40. The method of claim 31 , wherein the virtual agents are selected by the user according to a browsing model.
41. The method of claim 31 , wherein the group of selected virtual agents is selected according to a directory model, based on recognizing the context of the user request.
42. The method of claim 41 , further comprising the dialog manager generating a profile for each virtual agent, the profile comprising information or description regarding the content and services provided by the virtual agent.
43. The method of claim 42 , wherein the selected group of virtual agents is selected by evaluating and ranking the virtual agent based, at least in part, on applying a predetermined set of metrics on the recognized context.
44. The method of claim 43 , wherein the metrics are recall-oriented.
45. The method of claim 41 further comprising, prior to selecting the virtual agents, the dialog manager causing the conversational user interface to send an initial request to a predetermined set of candidate virtual agents to solicit a potential response.
46. The method of claim 45 , further comprising the dialog manager integrating the potential response using precision-oriented statistics.
47. The method of claim 46 , wherein the precision-oriented statistics comprise one or more of the following statistics: (a) inverse document frequencies, stop lists, historical requests, and context.
48. The method of claim 45 , further comprising providing a sample response database, wherein the dialog manager queries the sample response database with the initial request.
49. The method of claim 31 , further comprising grouping the virtual agents according to one or more of the following criteria: (i) conversation clustering; (ii) relevant conversation distribution; (iii) rule-based selection; (iv) conversation probing; and statistical.
50. The method of claim 31 , further comprising grouping the virtual agents based on one or more of the following methods: (i) relevant request distribution and (ii) content-based hots ranking and selection.
51. The method of claim 50 , wherein the relevant request distribution method comprises: (a) building a database using training data associates user requests with relevant virtual agents; (b) for each new user request: (i) finding in the database k nearest “neighbors”; and (ii) obtaining a normalized relevant conversation distribution using statistics associated with the k neighbors to obtain a relevance metric for each potential virtual agent; and (c) selecting a desired number of federated conversational agents from the relevance metrics.
52. The method of claim 31 , wherein the machine learning methods are used to adapt parameters of the dialog manager, which includes one or more of: domain, security, brand, past user responses, semantics, named entity features, category features, past user satisfaction features, vertical intent features, temporal features, text similarity features, hit count features, and other relevant parameters.
53. The method of claim 31 , further comprising selecting the group of virtual agents both according to the context and content of the user request.
54. The method of claim 31 , wherein the virtual agents are organized according to a message-passing model implemented in a peer-to-peer network.
55. The method of claim 54 , wherein each node of the peer-to-peer network comprises a client, a server, or both.
56. The method of claim 54 , wherein the peer-to-peer network implements a flood algorithm to distribute the user request to the virtual agents.
57. The method of claim 56 , wherein each virtual agent decides whether or not to respond to the user request.
58. The method of claim 54 , wherein the peer-to-peer network has a central-hub topology.
59. The method of claim 54 , wherein the peer-to-peer network has a hierarchical topology including hub nodes and leaf nodes.
60. The method of claim 54 , wherein the peer-to-peer network implements one or more of: (a) partial indexing over specific network resources; (b) structured peer-to-peer architecture using distributed hash tables, (iii) a gossip protocol, (iv) an architecture organized according to content or user interests; (v) super peers; and (vi) reputation management.
Cooperative Patent Classification codes for this invention. Click any code to explore related patents in that topic.
September 4, 2018
October 27, 2020
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.